I have no problems for 5 mK stability over long time with my crude setups and fancy Keithley 2510. Thermal sensors are either Honeywell HEL-705 RTD or YSI 44006.
Setup with ILX 5910B and YSI 44007 shows simialr performance too, with much smaller cost then 2510. 
10 mK stability over an hour period is enough for me, find some test results, temperature measure by Shimaden FP23 with a Pt100 RTD.
the Ambient temperature is around 15C, all the resistor under test are in the chamber expect SR-104.
TimB,
I would be really interested in hearing just a bit more about your project.
Those cheap coolers look like a perfect start for that.
You could always use two power op-amps in a bridge configuration for linear drive of a TEC as that’s pretty low noise. (Though I wouldn’t do that for more than a 5W TEC unless you have room for a large heatsink for the amps.)
That said, TI has a nice inexpensive H-Bridge that can be used for PWM based TEC drive up to 15A, with a couple of added inductors and filter capacitors. (So long as you keep the ripple below 10% you’re OK to drive a TEC with PWM in the hundreds of kHz range).
A third option would be to use “time proportional” control, coupled with a pair of relays, SSRs or any off the shelf H-Bridge. This is basically really, really slow PWM. Essentially, you set a time window (say 10 seconds) and turn the TEC fully on for a portion of that window (50% would be 5 sec on 5 sec off).
I’ve currently got a small TEC based cooler (designed to hold 6 cans of soda) that I converted to a temperature control chamber for testing voltage references. It uses the last option mentioned above. It’s got a range of 5c to 65c (at 25c ambient) and can maintain the setpoint +-0.25c, which is pretty good considering it uses nothing but parts from my junkbox.
Clearly this is an electronics forum, with many comments on circuit configuration, loop stability and the like, and only one comment on actual thermal modeling of the environment, plus Kleinstein's comment on heat distribution. Admittedly thermal modelling software is every bit as complicated as EDA software, and generally just as pricey so most of us have little access to it, but that isn't an excuse for ignoring key parts of the problem.
There needs to be attention paid to the fundamentals of the problem. Which elements (if any) need to be isothermal? Which need to be stable over time, but not necessarily isothermal? What are heat paths to and from those elements? What are the heat capacities and thermal conductivity of those pieces? What are the overall control loops, including those elements? Note that making heater power much greater than heat flow out makes control inherently non-linear and complicates stabilization. This is one advantage of TEC as you can get better balance to these flows.
For a reference oven, where there is much less time pressure to reach the operating temperature, no variation in the thermal characteristics of the controlled item, and little or no need to vary that temperature the inherent reliability of a mechanical solution (insulation, nested boxes etc) combined with a really simple outer control loop will be the best choice. Very different case than an environmental test oven.
To add some example to those questions I will add the info about patent 3,028,473 - Temperature stabilized oven, by George A. Dyer and Roy J. Patterson (Filed March 12, 1959) North American Aviation
'simple, highly reliable, and extremely compact device which is capable of regulating temperature within .01º F'
1x Thermistor 51A3 (Victor Engineering Company)
1x Thermistor D054 (General Electric Corp.)
1x NPN 690T1
1x NPN 2N547
1x Zener 911D18
1x Heater resistor 120 ohms
1x Heater resistor 30 ohms
1x Pot 25K
1x Pot 2K
1x Res 4.22K
I think it is an interesting (20 minutes, 4 pages) reading.
Might not be obvious to all, but the heater in the patent application uses cordwood packaging. Well suited to this purpose.
I'm currently working on a resistive (heating only) version of a similar box, which is intended solely for keeping components or a small board at exactly 25C.
It consists of 12 10R resistors affixed to the bottom of a Hammond 1590BB using Arctic Silver thermal epoxy, insulated with 1/4 inch "craft foam".
I hooked it up to a 12V supply (actually, at 95mA, to simulate 12V - 0.6V for a 2N3904, so about 1.1W) and let it run unregulated overnight, and it looks like it maxes out around 28 or 29C (in 72F ambient), so it should be able to regulate at 25C.
I hacked up a quick proof-of-concept on protoboard, which worked, but was noisy, and followed this up with a "proto shield" for Arduino. The results are okay: stable within about 0.05C (verified by using a Si7021 as the DUT)
Unfortunately there is a large measurement offset which occurs when the 12V power is switched on. I had attempted to create a bit of a star-ground, but there must be some current somewhere which is lifting up either the op amp or the MCP9701A. Nope, the op amp is being powered by the 12V rail, so naturally cutting its power will disrupt things...
Also, the ~1W max output turned out to be not quite sufficient. We had a cold front come through, and my apartment dropped from 72F to 67F overnight, and that was enough to cause the box to fall out of regulation.
Also, I haven't tuned the PID yet at all.
This is probably sufficient for what I needed it for, but I'd definitely like to improve upon it.
https://github.com/cellularmitosis/logs/blob/master/20180124-25c-chamber/README.mdedit: I suppose it is also possible the measurement offset is being caused by a magnetic field generated from the loop which the heater resistors form.
Here's the circuit
edit: forgot the cap on LM358 pin 6
Also, the ~1W max output turned out to be not quite sufficient.
no wonder:
I use usually 2 heater foils 12V/12W so 24 Ohms in series from a 17V unstabilized supply.
So I have 12W to heat up from room temperature (around 20 deg C) to the 30 deg C of my reference resistor heater.
(ok I need only around 20-25% of the power).
The 12W can go up only to around 50 deg C in my cooling box. (but that is sufficient for the 40 deg C that I need).
with best regards
Andreas
Search google for 'MCH'. It's a high efficiency heater.
I bought a 10x10mm MCH with 20 ohms or 40 ohms resistance (I bought two with different resistance and didn't remember which one was. In the worst case you can get +75C with only 1.25W).
This is the temperature test with PT100:
MCH 10x10 with PT100 : (ambient temperature around 20-22ºC)
@ 0.0V -> 108.62 ohms 22.11ºC
@ 0.1V -> 108.62 ohms 22.11ºC
@ 0.2V -> 108.63 ohms 22.13ºC
@ 0.3V -> 108.72 ohms 22.36ºC
@ 0.4V -> 108.82 ohms 22.62ºC
@ 0.5V -> 109.03 ohms 23.16ºC
@ 1.0V -> 110.63 ohms 27.29ºC
@ 1.5V -> 113.29 ohms 34.16ºC
@ 2.0V -> 116.58 ohms 42.67ºC
@ 2.5V -> 120.25 ohms 52.20ºC
@ 5.0V -> 142.10 ohms 109.48ºC
@ 7.5V -> 165.00 ohms 170.61ºC
Hello,
not bad.
I get similar performance for a PC-based PWM algorithm.
Switching every 1 second either on or off depending on the integrated measurement deviation.
And all via RS232 (RTS) signals with a simple FET.
for the 1 minute averaged values I see 0.012 K deviation (without noise) for a 2 deg C environment change.
within 1 minute the typical change is another 0.01K (0.35mV at NTC with 55mV / K sensitivity)
picture:
green: environment temperature (measured by ADC3 temperature sensor so contains 2-3 deg C self heating)
red: tempeature at reference resistor (setpoint 30 deg C). Lagging somewhat due to low amplification to avoid oscillation.
blue: heater power in % (from 12W).
With best regards
Andreas
Lady Heather has a temperature control pid routine that is used to stabilize a GPSDO environment. It uses two serial port modem control lines (for enable and heat/cool) to PWM a fan at 1Hz (could also drive a heater/cooler). The usual implementation is to mount the GPSDO in a box (usually with some added thermal mass... I use a simple cardboard box) and use a fan to blow air in. The temp control point is set below what the self-heating of the GPSDO is.
Attached is a plot of Heather stabilizing a Trimble Thunderbolt over 72 hours. It uses the temperrature value that the Thunderbolt reports. Over 72 hours the RMS error was below a microdegree. Instantaneous temperture change is in the low millidegree range.
I have an SR-104, a Thunderbolt, and a 10V ref installed in a box with the temp control point set to where the SR-104 is at exactly 10000 ohms. I also use the same algorithm on an ATMEGA-2561 touch screen micro to control the temp in my reflow toaster.
I worked up an LTSpice simulation using 10k thermistors rather than the MCP9701, and spun that up as an Arduino shield Kicad project. This design uses mostly "jelly-bean" parts, and also has an on-board thermistor for ambient temp. Note that the same resistor values work for a 5V supply as well.
github:
https://github.com/pepaslabs/isothermalOSHPark link:
https://oshpark.com/shared_projects/cw8b4XGm https://oshpark.com/shared_projects/9wWvXwMwWhen the boards arrive I'll solder it up and post here again.
I also ran my current prototype (using MCP9701) for 8 hours and it (mostly) stayed with +/-0.01C the entire time. Settings: kp = 10, ki = 0.2, kd = 0, period = 3000ms, setpoint = 25C (406 ADC counts).
I've also attached the current Arduino sketch.
Edit: Ah, shit. I accidentally swapped the position of the thermistor and fixed resistor. Luckily I can just invert the value of the ADC in software.
Edit2: attaching schematic.
So I thought about this a bit more and realized that the Arduino's 3.3V reference (and ADC) do not have a zero tempco.
So when the temperature of the Arduino changes, it effectively changes the temperature reading, which will introduce (undetected) error into the temperature control.
I epoxied an additional thermistor into the enclosure and hooked it up to my HP 34401A in 4-wire mode, and used a python script to convert from resistance to temperature.
I then logged the Arduino's notion of its error against the HP's notion of the error (see chart1.png). Turns out they disagree!
However, the smoking gun I need to is compare the difference in those two errors to the change in ambient temperature. If those are correlated well, that's evidence that the tempco of the Arduino directly causes additional undetected error in the temperature control loop. And a smoking gun it is indeed! (see chart2.png).
Edit: looks like the Arduino tempco introduces a control error of 0.02C per ambient C.
Edit2: (and that's assuming my HP 34401A has a zero tempco
So I thought about this a bit more and realized that the Arduino's 3.3V reference (and ADC) do not have a zero tempco.
If you have a ratiometric measurement then it should be only the drift of the amplifier.
Thats why I use my 24Bit ADCs also for temperature measurements.
(although the noise of local air drafts results in much less than 24 Bits).
In this case I have a full ratiometric measurement. And the drift of the ADC itself is negligible.
With best regards
Andreas
I'm currently working on a 10V heated reference totally homebred. It is based on the AD588KQ and is 'double' everything:
The power supply is linear and has double 78xx 79xx pairs to improve line regulation.
The whole enclosure is metallic and the AD588 will be enclosed in a small copper box too (double shielding)
Also the inside of the enclosure will be covered with heat insulating material, as well as the small copper box for the AD588 (double thermal insulation) (
https://uk.rs-online.com/web/p/thermal-insulating-sheets/7248909/)
Finally the temperature regulation is based on the TMP01 for its simplicity of use. A resistor switched to the mains rectified voltage serves as the heater. Also a DC fan running at low speed on the inside of the enclosure will try to keep the temperature constant.
Lady Heather has a temperature control pid routine that is used to stabilize a GPSDO environment. It uses two serial port modem control lines (for enable and heat/cool) to PWM a fan at 1Hz (could also drive a heater/cooler). The usual implementation is to mount the GPSDO in a box (usually with some added thermal mass... I use a simple cardboard box) and use a fan to blow air in. The temp control point is set below what the self-heating of the GPSDO is.
Attached is a plot of Heather stabilizing a Trimble Thunderbolt over 72 hours. It uses the temperrature value that the Thunderbolt reports. Over 72 hours the RMS error was below a microdegree. Instantaneous temperture change is in the low millidegree range.
I have an SR-104, a Thunderbolt, and a 10V ref installed in a box with the temp control point set to where the SR-104 is at exactly 10000 ohms. I also use the same algorithm on an ATMEGA-2561 touch screen micro to control the temp in my reflow toaster.
Thanks texaspyro, I just picked up a second 34401A and I’ll be trying the box + PWM fan method to see if I can give one of them a thermally regulated ambient environment, while I leave the second 34401a flapping the in breeze. Hopefully I’ll be able to see if my resistance measurement noise floor is due to the drift of my meter, or some other source.
Lady Heather has a temperature control pid routine that is used to stabilize a GPSDO environment. It uses two serial port modem control lines (for enable and heat/cool) to PWM a fan at 1Hz (could also drive a heater/cooler). The usual implementation is to mount the GPSDO in a box (usually with some added thermal mass... I use a simple cardboard box) and use a fan to blow air in. The temp control point is set below what the self-heating of the GPSDO is.
Attached is a plot of Heather stabilizing a Trimble Thunderbolt over 72 hours. It uses the temperrature value that the Thunderbolt reports. Over 72 hours the RMS error was below a microdegree. Instantaneous temperture change is in the low millidegree range.
I have an SR-104, a Thunderbolt, and a 10V ref installed in a box with the temp control point set to where the SR-104 is at exactly 10000 ohms. I also use the same algorithm on an ATMEGA-2561 touch screen micro to control the temp in my reflow toaster.
Thanks texaspyro, I just picked up a second 34401A and I’ll be trying the box + PWM fan method to see if I can give one of them a thermally regulated ambient environment, while I leave the second 34401a flapping the in breeze. Hopefully I’ll be able to see if my resistance measurement noise floor is due to the drift of my meter, or some other source.
First attempt: a small appliance light bulb inside of a cardboard box. There is an inlet hole and on the outlet there is a small fan affixed inside of a cardboard (toilet paper roll) tube. An arduino reads the temperature of an Si7021 (suspended in the middle of the box by another cardboard tube) and controls the duty cycle of the fan (on a ~400ms period).
Even with a crudely tuned PID loop, it was easy to hit +/-0.1C, in an environment which showed a 1C swing (measured outside of the box the day before).
Next, a box big enough to fit a 34401A into...
The first plot is the ambient temperature as measured the night and morning before this run. Second plot is the first 30-minutes (warmup) of the PID control, last plot is an all-day run.
(arduino code also attached)
edit: the fan is one of these 38mm units:
https://www.ebay.com/itm/222273911362 searching for "video card fan" seems to produce these "shroudless" fans which can be mounted inside of a tube.
(bonus analysis of the first plot, because it's fun
Finally getting back to this after a busy Summer.
Finished the rack ears and joiner for my 34970's today, 700W UPS fitted with a full load of new AGM's and a soon to be added temperature controlled fan (this model didn't have a fan), PC is a work in progress as are the extra Ref's and Batteries.
While this will be home to the References and my couple of Weston Cells and Resistors I am still playing with a portable stabilized enclosure which should be able to sit inside when at home. The more full the fridge is within reason the more stable it will be.
The plan is to keep the Rack Case in the region of 25C using heat scavenged off the UPS and other gear with a few fans if needed to minimise the power the fridge consumes so a small string of sensors placed around the Rack driving the fans as needed. As the Rack is going into a wheeled bench it can be moved around to help keep it cool or warm depending on the seasons.
The Fridge gets Ears tomorrow
The word "fridge" has been used a lot of times here, but without going all the way.
What you really want *is* a fridge, and then you want to do some serious case-modding.
The first thing to realize about temperature stabilization is that you need to think about both the energy-flow and the time constant.
Energy-flow first: How much electricity do you feed into your circuit? - You'll have to get rid of all of it again with your cooling system.
If you want stable temperature, the smartest thing you can do is to have stable energy-flow.
If you really want to go hard-core with this, you do not have anything which switches on/off, only things which send the same exact amount of energy in or or the other direction to keep the total energy-flow is constant.
Think in terms of running idle current through your calibration resistors also when you do not measure them, think in terms of Class-A amplifiers for anything regulated (ovens around Xtals, Rb's etc.)
Time-constant is a thing people almost always overlook: You can either go high or low.
If you go low time constant, you have as little (thermal) mass as possible in your controlled zone, and you use an aggressive airflow to keep the temperature stable.
This is far more tricky than it sounds, but is nice if you want to open the door and experiment with things. (This is why professional temperature chambers are expensive.)
If you go high time constant, you add (thermal) mass in your controlled zone, to help keep the temperature stable in time, just like you add big capacitors to keep voltages stable in time.
There are many ways to think about that time constant, there actually is something called "Thermal impedance" and you can get far by using Spice to simulate it.
So back to the fridge thing: If you want a cheap large enclosure with a stable temperature, find an old fridge.
Get rid of the compressor but first get somebody to remove the gas properly, we need that Ozone layer.
Now use the cooling circuit piping to circulate water instead, this allows you to remove quite a lot of heat from the interior, if you need to.
(You will want to replace the thin high pressure tubing, you cant get enough water through that.)
You probably want to control the inlet temperature of the water. Start with a variable speed fan on the external radiator and see where that gets you, upgrade to peltier if necessary.
Finally, if you do not plan to open the door all the time, add thermal mass *and* impedance. Some slices of cinderblock works great for that.
And obviously: Drill some holes for the test-wires, but remember copper is a very good thermal conductor.
Other hints: A small camera can save a lot of door-openings if you need to monitor indicator lights.
/Poul-Henning
Glad to see you on the forum Poul-Henning, I was wondering when you would make the jump from the volt-nuts mailing list